GxE for Genomic Prediction Models

Typically hybrid breeding programs identify elite cultivars by assessing the performance of large numbers of different combinations of inbred parent lines in multi-environment trials (METs). Computing possible dominance effects is particularly important for hybrid crops such as sorghum, where cultivars are hybrid combinations of female and male plant parents.

Raw plot yield was measured for sorghum hybrids in 2 sequential years of sorghum breeding trials, the first year had data from 7 sites and the second year contained 9 sites. These data were analysed as a 16 site MET with the hybrids fitted as random effects with the genetic variance partitioned into additive and dominance parts using genomic relationship matrices. The genomic relationship matrices used 30k SNP markers for 1124 hybrids, which were the combination of 565 male parents and 3 female parents.

Plant breeders commonly use a process of selection where the best hybrids from one year are advanced into trials the following year. However, genotype by environment interactions (GxE) were significant for both the additive and the dominance partitions of the genetic variance. While the additive effects are positively correlated for most sites, there were larger variations between sites for the dominance partition. The large degree of GxE in sorghum makes the across site predictions less meaningful for some environments. While the predictions of the additive effects for parents are usually quite meaningful across sites, predicting dominant hybrid effects are less meaningful.

These MET models using genomic relationships can predict hybrids that were not present in both years data. These genomic predictions from non-phenotyped sites may be compared with the predictions for the phenotyped sites. These results indicate that selection from a single year of trials may not be sufficient to produce superior hybrid lines for all environment types.